MCPcopy Create free account
hub / github.com/NVIDIA/cutlass / sync

Method sync

python/cutlass_cppgen/backend/arguments.py:102–118  ·  view source on GitHub ↗
(self, stream_sync=True)

Source from the content-addressed store, hash-verified

100 raise TypeError("Unsupported Frontend. Only support numpy and torch")
101
102 def sync(self, stream_sync=True):
103 if stream_sync:
104 (err,) = cudart.cudaDeviceSynchronize()
105 if err != cuda.CUresult.CUDA_SUCCESS:
106 raise RuntimeError("CUDA Error %s" % str(err))
107
108 for key in self.host_tensors.keys():
109 host_tensor = self.host_tensors[key]
110 (err,) = cuda.cuMemcpyDtoH(
111 host_tensor,
112 self.buffers[key].ptr,
113 host_tensor.size * host_tensor.itemsize,
114 )
115 if err != cuda.CUresult.CUDA_SUCCESS:
116 raise RuntimeError("CUDA Error %s" % str(err))
117
118 self.free()
119
120 def free(self):
121 """

Callers 1

runMethod · 0.45

Calls 1

freeMethod · 0.95

Tested by

no test coverage detected